The Agent Skills Directory

[DATA_EXFILTRATION]: The skill employs Puppeteer to capture screenshots of local files or URLs. These images are then uploaded to fal.ai's cloud storage using the @fal-ai/client. If the agent is instructed to process sensitive local files (such as .env files or configuration with hardcoded secrets), this mechanism will transmit visual representations of that sensitive data to a third-party API.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data (HTML and brand context) from the user's project.
Ingestion points: runtime/src/upgrade.mjs screenshots the target; runtime/src/implement.mjs and runtime/src/review.mjs ingest the original HTML source code.
Boundary markers: The skill uses BEGIN_HTML and END_HTML tags to delimit the ingested source code within the AI prompts.
Capability inventory: The skill possesses the ability to write files (writeFileSync), execute shell commands via FFmpeg (spawn), and perform network requests to external APIs (OpenRouter, fal.ai).
Sanitization: No sanitization or filtering is performed on the ingested HTML before it is sent to the LLM for analysis and redesign.
[COMMAND_EXECUTION]: The runtime/src/video.mjs script executes the ffmpeg binary using child_process.spawn to create design comparison videos. While the arguments are passed as an array, this grants the skill the ability to invoke system-level media processing tools.
[EXTERNAL_DOWNLOADS]: The skill interacts with external services including fal.ai and OpenRouter for design generation and vision-based analysis. It also references common CDNs like Tailwind and Google Fonts in its generated HTML output.

fal-redesign